Hands busy, voice free: capturing what the shift manager knows

The problem

The person with the most live information contributes the least to your systems — for physical reasons.

The person with the most live information in the whole plant is the one who contributes the least to your systems — not for lack of will, but for physical reasons. With hands busy and hearing protection on, writing is not an option, and nobody expects him to stop and do it. What is lost every day are precisely the early signals: that centrifugal 2 purges slower since the screen change, that the filter blinds earlier than normal, that the dryer runs wet since the raw sugar changed. Those are the ones that precede a stoppage or a quality deviation. That information dies at the end of the shift, or survives five minutes in a corridor conversation at handover. The handover report, written in a rush, captures a minimal fraction of it.

• With hands busy and hearing protection on, writing is not an option, and nobody expects him to stop and do it.
• What is lost every day are precisely the early signals: that centrifugal 2 purges slower since the screen change, that the filter blinds earlier than normal.
• Those signals are the raw material of any predictive maintenance plan.
• Without them, maintenance stays reactive and continuous improvement runs on memory.

How it fits the IRIS system

Connect in voice mode, with plant-rated hardware and its own wake word.

Connect in voice mode, with plant-rated hardware:

It works the other way round too: it returns context by voice — who has picked up the incident, when the last intervention on that equipment was, whether there is relevant history — without him going back to the office.

• A Bluetooth earpiece compatible with hearing protection, with its own wake word.
• The shift manager dictates while walking, in natural language.
• The model structures the dictation as an incident linked to the active equipment, lot and shift.
• iLEAN returns context by voice or push: who has picked up the incident, when the last intervention on that equipment was, whether there is relevant history. It does not decide for him: it gives him information so he decides better and faster.

See the full IRIS architecture →

Before and after

The shift manager's knowledge, before and after

Aspect	Today	With iLEAN Connect
Recording channel	Memory, and the handover	Continuous, during the round
Early signals recorded	None	Incidents tied to equipment and lot
Cost of recording one	Stop, remove protection, write	Dictate while walking
Looking up equipment history	Go back to the office	Ask and hear the answer
Hearing protection	Makes writing impossible	The earpiece is compatible with it
Shift handover	Whatever gets mentioned	What was recorded, with its context

Impact estimate

Impact estimate — an enabling piece, and it is worth saying so.

The block below is an estimate to be validated against your plant's actual data. We put it forward so the committee has an order of magnitude; we refine it during the assessment.

• This is the hardest piece to monetise on its own, and the brief says so openly rather than inventing a payback.
• Its value is enabling: every predictive maintenance and continuous improvement plan depends on having these signals recorded.
• Without them, the other cases measure the process but never capture why it drifted.
• It is defended as an enabler of the rest of the matrix, not with a number of its own.

this is the hardest piece to monetize on its own, and it is worth saying so openly. Its value is an enabling one: every predictive maintenance and continuous improvement plan depends on this record existing, because without it there is no data to feed them. *Estimate to be validated* with the continuous improvement lead.

And the fair question from the production manager

"Our shift managers speak in heavy plant slang." — all the better. Transcribing speech over a closed, known vocabulary — equipment, lots, symptoms — is an anchored task [1]. What gets structured are those entities, not free prose. And anything with low confidence is flagged for review rather than invented.

^[1] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.